I'm currently running zcs-NETWORK-5.0.6_GA_2313.RHEL5_64.20080522105815 on a AMD64 platform.
Every other day at different times, we have observed zmmailboxdmgr begin to consume memory and eventually starve the machine to such an extent that the swapping locks the machine with 45 load or until it becomes unresponsive. The only remedy at this point is a SysRq-E and reboot.
The process that causes this looks like this:
/opt/zimbra/libexec/zmmailboxdmgr start -Xms803m -Xmx803m -client -XX:NewRatio=2 -Djava.awt.headless=true -XX:MaxPermSize=128m -XX:SoftRefLRUPolicyMSPerMB=1 -Xss256k
The Zimbra server has 2 GB of physical RAM and 1 GB of swap. Under normal running circumstances (with Zimbra running), there is just under 1 GB free (uncached).
The last thing in mailbox.log is the following:
2008-06-11 22:15:11,163 INFO [btpool0-771] [firstname.lastname@example.org;mid=1;ip=22.214.171.124;] soap - NoOpRequest
2008-06-11 22:15:14,765 INFO [MailboxPurge] [email@example.com;mid=30;] purge - Purging messages.
2008-06-11 22:16:05,393 INFO [btpool0-771] [firstname.lastname@example.org;mid=1;ip=126.96.36.199;ua=Zimbra Desktop/0.84;] soap - SyncRequest
2008-06-11 22:16:11,680 INFO [btpool0-771] [email@example.com;mid=1;ip=188.8.131.52;] soap - NoOpRequest
2008-06-11 22:16:14,949 INFO [MailboxPurge] [firstname.lastname@example.org;mid=1;] purge - Purging messages.
I am currently using sun-jdk-1.6.0.06. The Zimbra documentation mentions that it should run on jdk 1.6. However, I'm wondering if it is safer to downgrade to jdk-184.108.40.206 or another jdk in the 1.5 series.
It's impossible to reproduce this issue manually (I've tried). However, it will inevitably happen again. The recovery isn't too disruptive, but we'd obviously like to fix it.
Please let me know if you have any advice and if you need additional information to diagnose the problem.
If the forums do not prove helpful, I plan to open a ticket with Zimbra support.